创新背景

量子物理学和微观领域研究的发展对于精密测量需求越来越大，相关技术需要借助人工神经网络进行深度学习增强传感。人工神经网络是受构成动物大脑的生物神经网络的启发生成的计算系统，它与生物神经网络不完全相同，通过例子来“学习”执行任务，而不用特定于任务的规则进行编程。

创新过程

《自然·通讯》在2022年4月14日发表了中国科技大学郭光灿院士团队利用里德堡原子和神经网络交叉在多频率微波传感上取得的新进展，成果论文题为《深度学习增强里德堡多频微波识别》。

里德堡态是指原子或分子的一种状态。在该状态下，原子或分子中的一个电子被激发到主量子数较高的轨道。里德堡原子具有超长寿命和对外场的高灵敏度，是处于高度激发态的原子。因为对外场的高灵敏度，里德堡原子具有较大的电偶极矩，敏感衡量正电荷分布与负电荷分布的分离状况，可以对微弱的电场产生很强的响应，对于精密测量领域尤其微波测量具有重要作用。微波测量是对工作于微波波段的元器件、电路、系统、传播媒质等性能与参量的测量，与微波通信的工作密切相关。

目前基于里德堡原子的微波测量还有许多不足之处亟需解决，比如多频率微波在原子中会引起复杂干扰，影响信号接受识别，妨碍多频率微波的接收。

研究将里德堡原子核神经网络相结合。先基于室温铷原子体系，让里德堡原子借助电磁诱导透明效应，检测多频微波场，发挥微波天线和调制解调器的作用，将接收到的调制信号通过神经网络的深度学习进行分析，高保真调解多频微波信号。经过神经网络深度学习后的里德堡微波接收器不需要其他复杂电路和带通滤波器，就可以一次直接解码20路频分复用（FDM）信号，准确率接近百分之百。

新方法利用高度里德堡原子的灵敏度优势进行微波测量，同时降低了噪声的影响。里德堡原子和神经网络交叉结合为精密测量领域提供了新的途径和方法，促进精密测量领域的发展以及神经网络学习功能在微观研究中的应用。

创新关键点

里德堡原子和神经网络交叉结合提高微波测量准确性。

Reedberg's cross-innovation of atoms and neural networks has led to the development of precision measurements

On April 14, 2022, "Nature Communications" published the new progress made by the team of Academician Guo Guangcan of the University of Science and Technology of China in multi-frequency microwave sensing using the intersection of Rydberg atoms and neural networks.”Deep learning enhanced Rydberg multifrequency microwave recognition.”
A Rydberg state refers to a state of an atom or molecule. In this state, an electron in an atom or molecule is excited to an orbital with a higher principal quantum number. Rydberg atoms have ultra-long lifetimes and high sensitivity to external fields, and are atoms in a highly excited state. Because of the high sensitivity to the external field, the Rydberg atom has a large electric dipole moment, which can sensitively measure the separation of the positive charge distribution and the negative charge distribution, and can produce a strong response to weak electric fields. Measurement plays an important role. Microwave measurement is the measurement of the performance and parameters of components, circuits, systems, and propagation media working in the microwave band, and is closely related to the work of microwave communication.
At present, there are still many deficiencies in the microwave measurement based on Rydberg atoms. For example, multi-frequency microwaves will cause complex interference in atoms, which will affect the reception and recognition of signals, and hinder the reception of multi-frequency microwaves.
The research combines a Rydberg nucleus neural network. First, based on the room temperature rubidium atomic system, let Rydberg atoms use the electromagnetically induced transparency effect to detect multi-frequency microwave fields, play the role of microwave antennas and modems, and analyze the received modulated signals through deep learning of neural networks, high-fidelity mediation Multi-frequency microwave signals. The Rydberg microwave receiver after deep learning by the neural network can directly decode 20 frequency division multiplexed (FDM) signals at a time without other complex circuits and band-pass filters, with an accuracy rate close to 100%.
The new method takes advantage of the sensitivity of highly Rydberg atoms to make microwave measurements while reducing the effects of noise. The cross-combination of Rydberg atoms and neural networks provides a new approach and method for the field of precision measurement, and promotes the development of the field of precision measurement and the application of neural network learning functions in microscopic research.